Server has 25 tools with consistent naming (verb_noun pattern) and descriptions present for all. However, critical gaps exist: (1) Input schemas lack proper JSON Schema type definitions for complex parameters (e.g., 'changes' object in update_tenant/update_user has no nested schema); (2) Parameter descriptions are sparse or missing constraints (e.g., 'canal' in trigger_sync lists valid values in description but not as enum); (3) Output schemas are completely undocumented, tools return JSON strings via _dump() with no schema guidance for LLMs; (4) Error handling is minimal, generic RuntimeError with HTTP status codes, no recovery guidance; (5) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. Naming is strong (whoami, list_tenants, create_tenant, trigger_sync are clear), but schema and output documentation are weak.

• Add JSON Schema definitions for object parameters. For update_tenant and update_user, define 'changes' as an object with properties: {nombre: string, plan: string, estado: string, fuente_tipo: string, notas: string, features: object} with descriptions for each field.
• Document output schemas for all tools. Specify the structure of returned JSON (e.g., list_tenants returns {tenants: [{id, nombre, plan, estado, fuente_tipo}], total: number}). Include field types and descriptions.
• Convert description-based enums to JSON Schema enums. For trigger_sync, change 'canal' parameter to include enum: ['all', 'shopify', 'mercadolibre', 'falabella', 'ripley', 'mundotec'].
• Add tool annotations via FastMCP. Use @mcp.tool(annotations={...}) to mark destructive tools with destructiveHint=true and read-only tools with readOnlyHint=true. This enables agents to reason about safety.
• Enhance error messages with recovery guidance. Instead of 'HTTP 401 No autorizado', return 'Authentication failed: token invalid or revoked. Check VAGO_AGENT_TOKEN environment variable and verify agent is enabled in Gestión de cuenta → Agentes IA.'